The grand vision of robotics for general-purpose machines that act, perceive, and learn in real-world environments must be met with intelligent capabilities of equal measure. To date, tremendous advances in robotics and artificial intelligence have been achieved primarily based on correlation. Yet, correlation is not causation: correlation-based methods often lack robust generalization, are brittle to distribution shifts, and can learn incorrect, spurious relationships in observed data. In contrast, a hallmark of human intelligence — reasoning about cause and effect — can provide clues for the next generation of robot intelligence. The advantages of adopting the principles of causality have been witnessed in fields such as biomedical science, economics, and genomics. Recently, it has been argued that the machine learning community should adopt the principles of causal inference towards causal learning of representations. Analogously, this workshop argues that the field of robotics stands to gain by integrating causality, moving robotics towards human-like embodied intelligence. Can robots learn and leverage the causal structure of problems? Can interventions and counterfactuals provide greater robot intelligence? And, ultimately, what next-generational capabilities can be unlocked through robots that can answer the question of “Why”?

8:50 - 9:00: Introduction

9:00 - 10:00: Session 1: Causal Learning and Representations

9:00 - 9:30: Peter Spirtes (CMU). Title: "Causal Learning Research: Achievements and Limitations"

9:30 - 10:00: Sara Magliacane (University of Amsterdam). Title: "Causal vs Causality-Inspired Representation Learning"

10:00 - 10:30: Break

10:30 - 11:30: Session 2: Foundations of Causality: Reasoning, Representations, Applications

10:30 - 11:00: Elias Bareinboim (Columbia). Title: "On the Causal Foundations of Artificial Intelligence (Explainability & Decision-Making)"

11:00 - 11:30: Francesco Locatello (ISTA). Title: "Foundations of Causal Representation Learning from Multiple Tasks and Sensors"

11:30 - 12:00: Rosemary Ke (DeepMind / Mila). Title: "Learning to Learn Causal Structure in Reinforcement Learning"

12:00 - 1:30: Lunch

1:30 - 2:30: Session 3: Reinforcement Learning and Representations

1:30 - 2:00: Biwei Huang (UCSD). Title: "Enhanced Reinforcement Learning through Causal Representations and Graph Structures"

2:00 - 2:30: Amy Zhang (UT Austin). Title: "Learning Invariant Representations for Out-of-Domain Generalization"

2:30 - 3:30: Poster Session

2:30 - 2:40: Lightning Talk (1 min per paper)

2:40 - 3:30: Poster Session

3:00 - 3:30: Break

3:30 - 4:30: Session 4: Causality and Robotics

3:30 - 4:00: Yunzhu Li (UIUC). Title: "Structural Causal Model Learning From and For Physical Interactions"

4:00 - 4:15: Ricardo Cannizzaro (Oxford). Title: "Causally-Informed Robot Planning in Partially-Observable and Confounded Environments"

4:15 - 4:30: Gokul Swamy (CMU). Title: "Causal Confounds in Imitation Learning"

4:30 - 5:00: Panel discussion: A "Fireside Chat" on Causality and Robotics with Biwei Huang and Yunzhu Li

5:00 - 5:05: Closing remarks

1. What Went Wrong? Closing the Sim-to-Real Gap via Differentiable Causal Discovery, Huang et al.

2. Towards Probabilistic Causal Discovery, Inference & Explanations for Autonomous Drones in Mine Surveying Tasks, Cannizzaro et al.

3. Identification of Average Treatment Effects in Confounded Additive Noise Models, Elahi et al.

4. Latent Space Planning for Unobserved Objects with Environment-Aware Relational Classifiers, Huang et al.

5. Object assembly using inferred causal models from humans, Tatlidil et al.

6. Embedding Scientific Knowledge via Visual Dirichlet Forests for Inference in Underwater Robotics, Samuelson et al.

7. Towards a Causal Probabilistic Framework for Prediction, Action-Selection & Explanations for Robot Block-Stacking Tasks, Cannizzaro et al.

8. Seeking Causal, Invariant Structures with Kernel Mean Embeddings in Haptic-Auditory Data from Tool-Surface Interaction, Khojasteh et al.

9. Learning Interpretable, High-Performing Policies for Continuous Control, Paleja et al.

10. Safety-aware Causal Representation for Trustworthy Reinforcement Learning in Autonomous Driving, Lin et al.

Broadly defined list of topics include, but are not limited to, the following

• Causal inference

• Sample-efficient robot learning

• Counterfactual reasoning in reinforcement learning

• Causal representation learning

• Causal discovery of latent graphs for robotic manipulation

• Generalized representation for robotic manipulation

• Physical scene understanding with causal predicates

• Interactive perception

• Experience-based causal skill discovery

• State space abstractions

• Learning from experts or oracle causal models

• Structured representations in robot perception and control

• Disentanglement and learning sparse mechanisms

• Causal transition models/dynamics and structured world models

• Attention, explainability, and interpretability through a causal lens

Important Dates: 

• Submission Deadline: September 1, 23:59 PT

• Decisions: September 8, 23:59 PT

• Camera Ready: September 25

• Workshop: October 5, 2023

Organizers

Sponsors